Closed-loop neurostimulation via reservoir computing

The idea of "brain controllability" refers to the possibility of manipulating brain activity through suitable external stimuli, chiefly for therapeutic aims (restoring normal brain activity in patients). Traditional approaches to dynamical system control generally require a system identifcation step where an explicit model of the system dynamics is obtained - a difficult step in the case of brain dynamics, where a plenitude of competing dynamical models exist. It was recently proposed that the system identification step could be circumvented by enacting control through a neural network (reservoir computer) coupled to the system. In this thesis, we will simulate brain activity with existing dynamical models and test the ability of the neural-network-based controller to modulate activity. Results may be of direct applicability to real neurostimulation experiments.